Open Access

ARTICLE

YOLO-CRD: A Lightweight Model for the Detection of Rice Diseases in Natural Environments

Rui Zhang^1,2, Tonghai Liu^1,2,*, Wenzheng Liu^1,2, Chaungchuang Yuan^1,2, Xiaoyue Seng^1,2, Tiantian Guo^1,2, Xue Wang^1,2

1 Tianjin Key Laboratory of Intelligent Breeding of Major Crops, Tianjin Agricultural University, Tianjin, 300392, China
2 College of Computer and Information Engineering, Tianjin Agricultural University, Tianjin, 300392, China

* Corresponding Author: Tonghai Liu. Email:

Phyton-International Journal of Experimental Botany 2024, 93(6), 1275-1296. https://doi.org/10.32604/phyton.2024.052397

Received 01 April 2024; Accepted 16 May 2024; Issue published 27 June 2024

Abstract

Rice diseases can adversely affect both the yield and quality of rice crops, leading to the increased use of pesticides and environmental pollution. Accurate detection of rice diseases in natural environments is crucial for both operational efficiency and quality assurance. Deep learning-based disease identification technologies have shown promise in automatically discerning disease types. However, effectively extracting early disease features in natural environments remains a challenging problem. To address this issue, this study proposes the YOLO-CRD method. This research selected images of common rice diseases, primarily bakanae disease, bacterial brown spot, leaf rice fever, and dry tip nematode disease, from Tianjin Xiaozhan. The proposed YOLO-CRD model enhanced the YOLOv5s network architecture with a Convolutional Channel Attention Module, Spatial Pyramid Pooling Cross-Stage Partial Channel module, and Ghost module. The former module improves attention across image channels and spatial dimensions, the middle module enhances model generalization, and the latter module reduces model size. To validate the feasibility and robustness of this method, the detection model achieved the following metrics on the test set: mean average precision of 90.2%, accuracy of 90.4%, F1-score of 88.0, and GFLOPS of 18.4. for the specific diseases, the mean average precision scores were 85.8% for bakanae disease, 93.5% for bacterial brown spot, 94% for leaf rice fever, and 87.4% for dry tip nematode disease. Case studies and comparative analyses verified the effectiveness and superiority of the proposed method. These research findings can be applied to rice disease detection, laying the groundwork for the development of automated rice disease detection equipment.

---

Keywords

---